Timely daily activity recognition from headmost sensor events

Smart homes are designed to promote safe and comfortable living for inhabitants without any manual intervention. The performance of approaches for daily activity recognition is therefore crucial, but current real-time approaches have to wait until a daily activity ends before performing recognition. We present an approach for timely daily activity recognition from an incomplete stream of sensor events, by which the recognition process can start as soon as a daily activity begins. Activity features are generated from several headmost sensor events rather than from all sensor events that a daily activity activated. A public dataset was utilized to evaluate the presented method. Experimental findings show its effectiveness for timely daily activity recognition in terms of precision, recall, average saved time, and saved time proportion. [Display omitted] [Display omitted] •A method for real-time daily activity recognition proposed.•The method only uses an incomplete stream of sensor events.•The proposed method can save time and attain comparatively good performance.